Working Notes: a commonplace notebook for recording & exploring ideas.
Home. Site Map. Subscribe. More at expLog. — Kunal

Ego

Ego is an Extensible Agent Orchestrator: inspired by emacs' architecture; I always feel compelled to hack on codex and claude, and I wanted something that would let me do that really trivially.

2026-02-22

General options

have multiple frontends, the way emacs does it
explore what I can achieve with different models
try to build support for recursive llm calls as a first take
AI can intercept any layer of this and rewrite it on demand, need to choose the right interaction model
At some point, try fine tuning models against this harness

The RLM paper has prompts in the appendix

explain that the context is available in a repl and how to control it
explain that python code to be executed must be specified with ```repl commands
give it a way to return the results using a function call or a variable
this is a whole tool ecosystem defined, I want to also extend this for it to be able to make and use tools automatically

I think next steps are to make a single process agent in python that can independently improve, and then figure out the human interaction model and instrumentation I'm planning to do.

2026-02-21

The general idea is to have a very simple python interpreter loop at the heart of the program that maintains state, and can easily be customized, hooked into, and overridden: exactly like emacs. I frequently feel like hacking on Claude Code, and then it's ... tricky? impossible?

The other bit is that I can have the AIs modify the orchestrator and save code in real time: and generally need to make sure this versionining is explicitly maintained so that restarting the interpreter maintains old state correctly. Particularly if the AIs want to make their own Python tools instead of installing shell commands.

I'd like to validate this with small experiments to see how far I can get, and also define an interface that works for this. Presumably users (& agents) will also want to install additional dependencies trivially into this application, so I need to figure that part out too.

Building a REPL

Shoutout to https://bernsteinbear.com/blog/simple-python-repl/ for a really good tutorial on building repls
Actual source code: https://github.com/python/cpython/blob/main/Lib/code.py

Simplifying API access

(via claude) https://rogs.me/2026/02/use-your-claude-max-subscription-as-an-api-with-cliproxyapi/
Based on the latest Tweets, this is Okay for local experiments: Thariq on Twitter
This should let me do far more interesting queries without having to deal with ACP or agent startup times